HubSpot’s Sidekick: Multi-Model AI Code Review with 90% Faster Feedback and 80% Engineer Approval

Editor's take

HubSpot has developed Sidekick, an internal AI tool that leverages multiple large language models to provide rapid, filtered feedback on code pull requests. This initiative addresses the persistent bottleneck of code review, aiming to significantly reduce developer wait times and improve the quality of submitted code. By automating initial checks and prioritizing actionable suggestions, Sidekick could free up human reviewers to focus on more complex architectural and design considerations.

The impact of such tools extends beyond individual engineering teams. Faster feedback loops are crucial for maintaining development velocity, especially as companies like HubSpot scale their operations and deploy more sophisticated AI-powered products. The reported 90% reduction in feedback time and 80% engineer approval rate suggest a practical, albeit internal, validation of AI's role in improving developer workflows. This could set a precedent for other organizations grappling with similar review inefficiencies.

Future developments will likely focus on Sidekick's adaptability to different programming languages and its integration with existing CI/CD pipelines. A key question will be whether the "sec" (presumably security or second-level) filtering mechanism effectively catches nuanced bugs or introduces its own biases. Observing how Sidekick's performance metrics hold up over time and if it can be generalized to external use cases will be telling.